摘要
提出了一种基于离散余弦变换的红外目标特征描述方法,通过离散余弦变换描述子提取平移、旋转、尺度放缩等变换下不变性的目标形状特征,利用前向传播神经网络学习及识别这些特征向量,对3类不同姿态下的目标图像进行了分类识别,并将所提出的特征提取方法与傅立叶描述子进行比较。仿真结果表明,和傅立叶描述子相比,在保证识别率不变的情况下,采用余弦变换描述法可以减少数据运算量和运算时间。
The discrete-cosine-transformation-based features of infrared targets are proposed, and the invariant features to translation, rotation, and scale change of targets are taken by discrete cosine transformation descriptors. The BP neural network is trained using the invariant features taken, and then recognition ability of the network is tested. Feature describing ability of the discrete cosine transformation descriptor is compared with the Fourier descriptor through recognizing three sorts of targets images in different poses by BP network trained. Recognition result shows that quantity computed and time expended are reduced obviously using the discrete cosine transformation based algorithm in the same recognition rate compared with the Fourier descriptors.
出处
《系统仿真学报》
EI
CAS
CSCD
北大核心
2005年第6期1363-1365,1369,共4页
Journal of System Simulation
关键词
离散余弦变换
边缘特征
神经网络
目标识别
discrete cosine transformation
boundary feature
neural network
target recognition